Memory Command Bus Training Using Parity-Based CA Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing command bus training methods for LPDDR4 and LPDDR5 memory devices are slow and require significant firmware coordination, limiting the speed and quality of command bus training due to the need for manual sampling and error checking of CA patterns, which can miss thousands of potential sampling points.
Innovation Solution
Implementing a command address training mode (CATM) similar to that used in DDR5 memory devices, which simplifies error checking by using parity values to indicate errors, reducing the need for firmware coordination and speeding up the command bus training process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional command bus training methods are used for LPDDR4 and LPDDR5 memory devices, then firmware coordination and manual sampling can ensure error checking, but the training process becomes slow and requires significant firmware coordination
Solution Approach 1:
The patent replaces the mechanical firmware coordination system with an automated command address training mode (CATM) system. The memory device autonomously samples CA patterns and generates parity values without requiring firmware intervention for each sampling operation, substituting manual firmware control with self-contained device automation.
Solution Approach 2:
The memory device performs self-service by autonomously sampling CA patterns, compressing them using XOR circuitry to generate parity values, and returning results to the memory controller without external firmware coordination. The device serves itself by independently completing the entire training measurement process.
2Measurement precision
If manual sampling and error checking of CA patterns is performed, then error detection can be achieved, but thousands of potential sampling points are missed
Solution Approach 1:
The CATM enables continuous sampling of CA patterns at all potential sampling points without interruption. The memory device continuously samples and processes CA patterns throughout the training window, ensuring no sampling points are missed while maintaining constant measurement action without firmware intervention between samples.
Solution Approach 2:
The patent accelerates the sampling process by using parallel XOR circuitry to rapidly compress multiple CA pattern samples simultaneously. The strong computational capability of the XOR-based compression circuit enables fast processing of thousands of sampling points in parallel, dramatically reducing the time required compared to sequential manual sampling.
3Reliability
If firmware coordination is used for each sampling operation, then error checking can be performed, but the complexity of firmware coordination increases
Solution Approach 1:
The patent extracts the error checking and sampling coordination functions from the firmware and relocates them to the memory device itself. The complex firmware coordination logic is removed and replaced with simple memory controller commands, while the memory device assumes responsibility for autonomous error detection and sampling management.
Solution Approach 2:
The patent introduces parity values as an intermediary representation of CA pattern sampling results. Instead of requiring firmware to directly interpret and check raw CA patterns, the memory device compresses patterns into simplified parity values that serve as intermediaries, making error detection straightforward without complex firmware logic.
4Manufacturing precision
If traditional training methods are used, then voltage and timing settings can be adjusted, but the process requires significant time and effort
Solution Approach 1:
The memory device performs preliminary sampling and parity generation for all potential voltage and timing settings before the memory controller needs to make adjustments. By pre-compressing CA patterns and generating parity values for multiple settings in advance, the device eliminates the need for time-consuming sequential adjustment and measurement cycles.
Solution Approach 2:
The training system dynamically adapts by using the compressed parity values to rapidly determine optimal voltage and timing settings. The memory controller can quickly adjust settings based on real-time parity feedback without waiting for lengthy measurement cycles, enabling dynamic optimization of training parameters throughout the process.
Data Source
AI summary
Techniques for command bus training to a memory device includes triggering a memory device to enter a first or a second command bus training mode, outputting a command/address (CA) pattern via a command bus and compressing a sampled CA pattern returned from the memory device based on whether the memory device was triggered to be in the first or the second command bus training mode.